Implement (most of) the rest of redis.* properties and methods for Lua scripts

[kevin-montrose]

Knocking these out now, as refactoring to work around longjmp is going to be onerous and tricky - and I want a good set of tests for all functionality before doing that.

Adds:

• redis.pcall
• redis.sha1hex
• redis.LOG_DEBUG, redis.LOG_VERBOSE, redis.LOG_NOTICE, redis.LOG_WARNING
• redis.log - along with new config LuaLoggingMode
• redis.set_repl - this raises an unconditional error, as it doesn't make much sense in Garnet
• redis.replicate_commands - this unconditionally returns true, as in Redis 7+
• redis.breakpoint and redis.debug - these raise unconditional errors, as implementing them is both weird and a big ask
• redis.acl_check_cmd - the lion's share of the PR, see below
• redis.setresp - also a chunky part of the PR, see below
• REDIS_VERSION and REDIS_VERSION_NUM
• Global variables forbidden
• Global functions forbidden

I'll do a separate change for the runtime libraries not currently supported, as I also think we want some ability to configure what's actually imported.

---

Details

redis.log

Mapping between Redis log levels and .NET log levels is:

• LOG_DEBUG -> Debug
• LOG_VERBOSE -> Info
• LOG_NOTICE -> Warning
• LOG_ERROR -> Error

By default redis.log(...) writes to whatever ILogger is configured. LuaLoggingMode can be used to configure as: Enable (default), Silent (call validated, but logs not written), or Disable (call validated, and then error raised).

Silent may seem strange, but is meant for the case where Garnet is hosted as a service which doesn't want to pollute its logs with user logs normally, but wants the option to gather those logs. Silent allows log lines to be placed, but ignored most of the time.

redis.acl_check_cmd

This method checks if the current user has permission to run a RESP command.

Trick is, you don't have to pass the whole command (in the sense of all arguments, or even subcommands). Most of our code assumes we have a whole command available to parse or otherwise validate.

So this implementations fakes it. It consults RespCommandsInfo just enough to figure out minimum argument counts (I had to fix a few incorrect Arity values in there), and creates a dummy RESP request array to shove into RespServerSession.ParseCommand via a new helper.

Some small changes were necessary to make all that work. I changed the ACL tests to also exercise this method, to give us more confidence in correctness.

redis.setresp

This method changes how LUA <-> RESP values are mapped depending on RESP version. This affects both the results of scripts, and the results of redis.(p)call, so it's rather involved.

As part of testing this, I fixed a few commands (HGETALL, SDIFF, and ZSCORE) whose response types change under RESP3 to correctly return new types. I strongly suspect there are a lot more commands that aren't accounting for RESP3, but will do that work in a separate PR.

Forbidden global variable and function declarations

In main we allow global declarations, which can carry over between script invocations. Redis forbids this, and it's a giant footgun - these globals are only present in a particular session, and their lifetimes are heavily implementation dependent.

These are now forbidden, which is technically a breaking change. Anyone depending on this behavior was already subtly broken, so I think it's fine.

Note that Redis appears to implement this by patching Lua - I did it in pure Lua, which imposes higher overhead but lets us keep using vanilla Lua 5.4.

---

Benchmarks

Since this change adds a fair amount to the bootstrapping of a Lua VM, I expected some loss in initial script loading - which is observed. Preventing global functions and variables had the risk of a general regression in Lua performance, but that does not seem to have happened.

All benchmarks are as of 77c1587bfa894e1884b2e5275ee3e798b1e3b9c8 for redisLuaParity, and 0c8bd6ffffa26b3b0cfd3bceaa5afa0890fae79c for main. I'm including just the Native,None results for conciseness.

LuaRunnerOperations

As expected, we're giving up a bit in construction (when the VM is created) and compilation (when various global functions are initialized). Worst case loss of ~40% (for ConstructSmall) - some of this is simply unavoidable, and this is typically amoratized over multiple script invocations.

main

Method	Params	Mean	Error	StdDev	Median	Allocated
ResetParametersSmall	Native,None	6.256 us	0.4919 us	1.427 us	6.100 us	352 B
ResetParametersLarge	Native,None	6.204 us	0.5221 us	1.515 us	5.700 us	448 B
ConstructSmall	Native,None	201.292 us	5.0428 us	13.889 us	198.500 us	1008 B
ConstructLarge	Native,None	217.891 us	8.8451 us	25.802 us	209.100 us	3448 B
CompileForSessionSmall	Native,None	32.873 us	1.4644 us	4.178 us	32.850 us	1024 B
CompileForSessionLarge	Native,None	119.605 us	6.9591 us	20.410 us	120.950 us	688 B

redisLuaParity

Method	Params	Mean	Error	StdDev	Median	Allocated
ResetParametersSmall	Native,None	7.269 us	0.5090 us	1.477 us	7.650 us	688 B
ResetParametersLarge	Native,None	6.689 us	0.4603 us	1.343 us	6.800 us	1024 B
ConstructSmall	Native,None	280.155 us	8.1001 us	23.110 us	277.400 us	504 B
ConstructLarge	Native,None	293.043 us	7.6313 us	21.896 us	292.800 us	3616 B
CompileForSessionSmall	Native,None	38.793 us	1.9871 us	5.605 us	38.100 us	352 B
CompileForSessionLarge	Native,None	121.113 us	6.0109 us	17.629 us	114.600 us	736 B

LuaScripts

Some jitter here, but no real change.

main

Method	Params	Mean	Error	StdDev	Gen0	Allocated
Script1	Native,None	124.4 ns	0.99 ns	0.87 ns	-	-
Script2	Native,None	152.7 ns	0.52 ns	0.43 ns	0.0002	24 B
Script3	Native,None	235.1 ns	2.25 ns	1.88 ns	0.0005	32 B
Script4	Native,None	243.0 ns	2.53 ns	2.36 ns	-	-

redisLuaParity

Method	Params	Mean	Error	StdDev	Gen0	Allocated
Script1	Native,None	132.3 ns	1.72 ns	1.52 ns	-	-
Script2	Native,None	159.4 ns	2.16 ns	2.02 ns	0.0002	24 B
Script3	Native,None	236.8 ns	1.86 ns	1.74 ns	0.0005	32 B
Script4	Native,None	221.3 ns	1.39 ns	1.30 ns	-	-

ScriptOperations

Some jitter here as well, but not real changes.

main

Method	Params	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
ScriptLoad	Native,None	82.36 us	1.578 us	2.107 us	-	-	-	9600 B
ScriptExistsTrue	Native,None	18.85 us	0.187 us	0.175 us	-	-	-	-
ScriptExistsFalse	Native,None	17.52 us	0.108 us	0.096 us	-	-	-	-
Eval	Native,None	61.40 us	0.285 us	0.267 us	-	-	-	-
EvalSha	Native,None	26.38 us	0.200 us	0.167 us	-	-	-	-
SmallScript	Native,None	54.09 us	0.254 us	0.198 us	-	-	-	-
LargeScript	Native,None	4,117.96 us	58.913 us	55.107 us	-	-	-	5 B
ArrayReturn	Native,None	114.84 us	2.295 us	2.732 us	-	-	-	-

redisLuaParity

Method	Params	Mean	Error	StdDev	Median	Gen0	Gen1	Gen2	Allocated
ScriptLoad	Native,None	78.79 us	0.278 us	0.247 us	78.87 us	-	-	-	9600 B
ScriptExistsTrue	Native,None	19.69 us	0.235 us	0.220 us	19.70 us	-	-	-	-
ScriptExistsFalse	Native,None	17.72 us	0.258 us	0.241 us	17.60 us	-	-	-	-
Eval	Native,None	62.02 us	0.392 us	0.348 us	62.08 us	-	-	-	-
EvalSha	Native,None	27.17 us	0.192 us	0.170 us	27.09 us	-	-	-	-
SmallScript	Native,None	52.28 us	0.410 us	0.383 us	52.18 us	-	-	-	-
LargeScript	Native,None	4,010.61 us	45.194 us	40.064 us	4,008.89 us	-	-	-	-
ArrayReturn	Native,None	116.23 us	0.729 us	0.609 us	116.00 us	-	-	-	-

[badrishc]

LGTM overall, some minor comments.